Analysis and numerics of the chemical master equation

Abstract

It is well known that many realistic mathematical models of biological and chemical systems, such as enzyme cascades and gene regulatory networks, need to include stochasticity. These systems can be described as Markov processes and are modelled using the Chemical Master Equation (CME). The CME is a differential difference equation (continuous in time and discrete in the state space) for the probability of a certain state at a given time. The state space is the population count of species in the system. A successful method for computing the CME is the Finite State Projection Method (FSP). The purpose of this literature is to provide methods to help enhance the computation speed of the CME. We introduce an extension to the FSP method called the Optimal Finite State Projection method (OFSP). The OFSP method keeps the support of the approximation close to the smallest theoretical size, which in turn reduces the computation complexity and increases speed-up. We then introduce the Parallel Finite State Projection method (PFSP), a method to distribute the computation of the CME over multiple cores, to allow the computation of systems with a large CME support. Finally, a method for estimating the support a priori is introduced, called the Gated One Reaction Domain Expansion (GORDE). GORDE is the first domain selection method in the CME literature which can guarantee that the support proposed by the method will give the desired FSP approximation error. To prove the accuracy and convergence of these three methods, we explore non-linear approximation theory and the theory of Chemical Master Equations via reaction counts. Using these tools, the proofs of the accuracy and convergence of the three methods are given. Some numerical implementations of the three methods are given to demonstrate experimental speed-up in computing an approximation of the CME.